In this chapter, the joint design of VMIMO technique and a mobile sink for data gathering in WSNs is proposed. The energy consumption when communicat- ing with SISO and VMIMO (both DSTBC and VBLAST) is firstly studied. Our studies demonstrate that utilizing VMIMO (either DSTBC or VBLAST) requires less overall energy consumption - both the transmission energy consumption and circuit energy consumption - than using SISO when communicating one bit data when the transmission distance setting as more than 10m. The experimental results illustrate the promising benefit from the utilizations of VMIMO. For the communication with distance more than 20m, DSTBC is more energy efficient compared to VBLAST due to that the increase in the transmission energy con- sumption of VBLAST outweighs the reduction in circuits and cooperation energy consumptions. By exploring the trade-off between the minimum network energy consumption (the fully utilization of VMIMO) and minimum data gathering la- tency (the shortest moving tour), the DAEE problem is formulated into an integer linear program and propose the WR algorithm to solve it.
WR combines both energy and latency revenues in its weighted revenue metric for choosing polling points. In doing so, it exhibits a good adaptivity to different network scenarios. Extensive simulation results on randomly deployed large-scale networks demonstrate the effectiveness of the proposed algorithm. Specifically, WR reduces the overall network energy consumption with limited moving tour length. The results also show that WR controls its emphasis aggressiveness and can be adaptively applied for different QoS-requirement applications by adjusting the weighting factors.
97
Chapter 5
Time Efficient Data Collection
with Mobile Sink and VMIMO
Technique
Even though mobile sink achieves uniform distribution of energy consumption, it comes with cost and introduces sink moving delay during data collection. On the other hand, by allowing concurrent uploading of different independent data streams, the overall data uploading time can be largely reduced in VMIMO com- munication networks. In this chapter, the overall data collection latency including both data uploading time and mobile sink moving time is considered, aiming to achieve the trade-off between the full utilization of concurrent data uploading and the shortest sink moving tour.
5.1
System model and problem formulation
Mobile sinks in WSNs alleviate hot-spot problems and helps to achieve uniformity of energy consumption in networks. However, as the sink moves, the data has to be buffered in the sensor nodes to wait for the arrival of the mobile sink, which
introduces sink moving delay and increases the total data gathering latency. In addition to the mobile sink moving time, the data uploading time also needs to be considered in the total data collection time. For some environment monitoring and data sensing scenarios, the amount of sensing data could be large, as well as the number of applied sensors. In both cases, the WSNs generate huge amount of total uploading sensing data. Due to the limited wireless effective transmission rate, the data uploading time could be longer that the MS moving time and even dominates the total data collection latency.
As is shown in Fig. 4.3, both Vertical Bell Laboratories Layered Space Time (VBLAST) and Distributed Space Time Block Code (DSTBC) can be used to achieve diversity gain and conserve energy, and at the maximum diversity gain, DSTBC requires less energy consumption for transmitting one bit data than VBLAST. However, to achieve multiplexing gain, VBLAST can offer a higher transmission rate, ideally M folds, where M is the minimum number of transmit antennas and the number of receive antennas [42–44]. In this case, independent data streams are allowed to be uploaded concurrently and the time saving could be significant with a huge advantage. The transmission time is then 1/M that of DSTBC, leading VBLAST a promising solution for delay-sensitive and energy constraint high data rate WSNs. The multiplexing mode of VMIMO is considered in this work. Hence, while taking advantage of energy efficient properties, delay minimization problem should be considered for those delay-sensitive applications in combined mobile sink and VMIMO communication system.
As mentioned, in some environment monitoring and military scenarios, the num- ber of sensors and the sensing data could be large enough so that the data up- loading time may largely affect or dominate the total data collection time [39]. The most worthy information comes from the time-sensitive data and the data collection delay could be of vital importance. For example, in military defence applications, sensors deployed in reconnaissance missions need to transmit back
5.1. System model and problem formulation 99
high-definition images and audio/video recording to identify hostile units. Delays in gathering sensed data may not only expose sensors or mobile sink to enemy surveillance, but also depreciate the time value of gathered intelligence. Using VMIMO can greatly speed up data collection and reduce overall latency [41]. Therefore, the data collection latency minimization problem is an important task in both research and practical applications.
In some practical scenarios, it is impossible to obtain pre-knowledge about the area and location information. Thus, without loss of generality, the polling point information is not pre-knowledgeable in this work. Given a set of sensors ran- domly deployed in the field, the polling points are selected from the sensors. The polling points can be part of the compatible pairs, and also the non-compatible sensors. The sensors that are associated with the compatible sensors or PPs are called association sensors. The association can happen by multiple hops. Once selected as a polling point, in addition to deliver its own sensing data, the sensor is also responsible for aggregating, buffering and transferring data from its asso- ciated sensors to the mobile sink. Therefore, there are different ways for sensing data to be collected by the sink:
(i). Compatible sensors upload their sensing data to the mobile sink concur- rently and directly when they are within the cover range of mobile sink. (ii). Association sensors send their sensing data to the associated compatible
sensors and polling points to buffer possibly via multihop. Upon the arrival of the mobile sink, the polling points and other compatible sensors upload their buffered data by VMIMO or SISO communications.
(iii). Association sensors that are associated with the non-compatible polling points can upload their sensing data directly to the mobile sink by one hop SISO communication when the mobile sink arrives within transmission range.
e a b c d f g h k m (a) a b c d e f g h k m (b) a b c d e f g h k m Sensor node Polling point Wireless link Compatible pair MS moving tour (c)
Figure 5.1: Three possible movement patterns for a mobile sink.
Fig. 5.1 shows three possible association patterns with different compatible pairs and corresponding two moving tours of the MS. In Fig. 5.1(a) two sensors (a and d) are selected as PPs and three compatible pairs are formed among the sensors (a, b), (c, d), and (e, f ). Sensor nodes h, g and k are associated with a, d and f respectively by one hop distance. Sensor node m is associated with a by two hops via h. In Fig. 5.1(b), three are four compatible pairs formed during the MS tour, and three sensor nodes (b, c and g) are selected as PPs. In Fig. 5.1(c), three compatible pairs are formed with three PPs are selected. Thus, for the three cases, case (a) selects the minimum number of PPs and could get the shortest sink moving time. Case (b) forms the maximum number of compatible sensors with three PPs. It gets longer moving time, but it achieves more concurrent uploading benefit which leads to less data uploading time. Even though case (c) forms three compatible pairs which is less than that in case (b), but it may achieve less overall data uploading time. This is attributed to that the different amount of data for the two compatible sensors has to be uploaded in SISO way in case (b). While in case (c) sensors h and g buffer and upload the same amount of data from its associated sensors. That is to say, the sensing data from association sensors d, m, f and k can be also uploaded to MS benefitting the concurrent data uploading via the compatible pair (h, g). All the sensing data in this case can be uploaded concurrently, hence, it achieves high utilization of VMIMO and small
5.1. System model and problem formulation 101
data total uploading latency.
Therefore, to achieve the minimum total data collection delay does not necessar- ily mean to form the maximum number of compatible pairs or to establish the shortest moving tour, it should also consider the amount of concurrently uploaded data. Hence, how to jointly utilize VMIMO and organize the selection of polling points to achieve the minimum total data collection latency is challenging. In this section, we study the Delay Minimization For Multihop Data Collection (DM- MDC) problem. Our objective is to minimize the total data collection latency. With regard to the trade-off between sink moving time and data uploading time, the optimal solution results may not achieve the shortest sink moving time or the shortest data uploading time.
However, multihop transmission costs more energy for delivering sensing data to the sink. In order to limit the energy consumption for sensor nodes, the maxi- mum number of hop distance can be bounded. Due to the technical limitation of sensors, the buffer size can also be bounded. VBLAST based VMIMO com- munication can be achieved easily with regard to the timing synchronism among sensors [43]. The synchronization of sensors can be done when the mobile sink arrives at the polling points. Mobile sink broadcasts its advertisement and all the sensors synchronise their clocks for time synchronization [121]. CSI is assumed to be perfectly knowledgeable to the receiver [129]. The cost of sharing control information for VMIMO transmission in the data gathering is not considered. The reason is that the control packet is relatively short compared with the data packet and thus the energy consumption of additional data exchange will not greatly impact the energy consumption of VMIMO communication [39].
Given a set of sensors S = 1, 2, ...N deployed over a sensing field, the DMMDC problem is to determine the selection of polling points, the compatible pairs, and the multihop associations between sensors to achieve the minimum data collection delay for sensors. Due to the limited resources of sensors, the buffer size of each
sensor is bounded with B and the maximum distance for multihop transmission is bounded with H. The amount of sensing data for each sensor in one data collection cycle is R (R < B). For a clear presentation, the notations used in the formulation are summarised in Table. 5.1.
The total data collection latency minimization problem can be formulated as:
min xmih,uij,eij {N − |S| X i=1 |S| X j=1 [uij + min( |S| X m=1 H X h=1 xmih, |S| X n=1 H X h=1 xnjh) · uij]/2} · R/Vr + |S| X i=1 |S| X j=1 Dij· eij/Vm (5.1) subject to |S| X m=1 H X h=1 xmih· R ≤ B (i = 1, ..., |S|) (5.2) |S| X i=1 H X h=1 xmih+ |S| X j=1 umj + km = 1 (i = 1, ..., |S|) (5.3) xmih ≤ |S| X j=1 uij + ki (i = 1, ..., |S|, h = 1, ..., H) (5.4) |S| X i=1 eij = ai (i = 1, ..., |S|) (5.5) |S| X j=1 eij = aj (i = 1, ..., |S|) (5.6)
Given the notation in Table. 5.1, the DMMDC problem has been formulated as an integer linear program labelled from Eqn. (5.1) to Eqn. (5.6).
The objective function Eqn. (5.1) minimizes the total data collection latency which includes both data uploading time and MS moving time. The part of min(P|S| m=1 PH h=1xmih, P|S| n=1 PH
h=1xnjh) specifies that the data collection time
5.1. System model and problem formulation 103
Table 5.1: Formulation notations. Indices:
S = {Si; (i = 1, ..., N )} A set of sensors.
Constants:
R ≥ 0 The amount of sensing data for each sensor.
Dij ≥ 0 Distance between two sensors Si, Sj.
H > 1 Maximum hop boundary for multihop transmission.
B > 0 Buffer size of each sensor.
Vm> 0 Velocity of mobile sink.
Vr> 0 Effective data uploading rate.
Variables:
ai = {0, 1} If sensor Siis selected as a polling point, ai= 1, otherwise,
ai= 0.
ki= {0, 1} If a polling point Si is non-compatible, ki = 1, otherwise
(PP i is part of compatible pairs), ki = 0.
uij = {0, 1} If the sensors Si and Sj are formed as a compatible pair,
uij= 1, otherwise, uij = 0.
eij = {0, 1} If the moving tour contains the segment between Siand Sj,
eij = 1, otherwise, eij= 0.
xmih= {0, 1} If the sensor Smis associated with sensor Si in h hop dis-
tance, xmih= 1, otherwise, xmih= 0.
Constraint (5.2) guarantees that the overall buffering data in any sensor is not exceeding the sensor buffer limit.
Constraints (5.3) - (5.4) guarantee that each sensor should be formed as part of the compatible pairs or be selected as non-compatible PP or be associated with one of them, so that its sensing data can be collected during the moving tour. Constraints (5.5) - (5.6) guarantee that the mobile sink enters and departs each polling point only once.
The objective of the problem is to find a tour and the association relations between sensors, such that (i) all the sensors are formed as a compatible sensor or selected as a PP or associated with one of them, (ii) the total data collection time for each sensor is minimized, (iii) the total buffering data and the maximum hop distance are within sensors’ constraints. DMMDC is NP-hard and it can be shown by a reduction from the well-known TSP problem. The total data collection time includes overall data uploading time which is affected by the total amount of concurrent-uploading data, and sink moving time which depends on the MS
moving tour length. In a special case where the network is super sparse. Assume that the sensing area is sufficiently large that no two sensors are able to form compatible pairs and no sensor can be associated with other sensors. In this case, all the sensors have to be selected as polling points and the mobile sink visits all of them. Since the amount of sensing data are the same for all sensors, the overall data uploading time is proportional to the number of sensors. Thus, to achieve the minimum overall data collection time, the moving time for mobile sink should be minimized. Hence, the solution is to find the optimal shortest moving tour to visit all the sensors once, which forms a minimum distance TSP problem. Hence, DMMDC is NP-hard.